Mark Gordon PowerPoint Presentation, PPT - DocSlides

Mark Gordon

1

COMET: Code Offload by

Migrating Execution Transparently

OSDI'12

Mark Gordon, Anoushe Jamshidi,Scott Mahlke, Z. Morley Mao, and Xu Chen

University of Michigan, AT&T Labs - Research

Mark Gordon

2

Overview

Introduction

Distributed Shared Memory

COMET Design

Evaluation

Summary

Mark Gordon

3

What is offloading?

Mobile devicesHave limited resourcesAre well connectedCan we bring network resources to mobile?Can a system transparently make this available?

Mark Gordon

4

Related Work

MAUI and CloneCloud

Utilize server resources

Computation, energy, memory, disk

'Capture and migrate' method level offloading

Areas for improvement

Thread and synchronization support

Offload part of methods

Mark Gordon

5

COMET's Goals

Improve mobile computation speed Require no programmer effort Generalize well with existing applications Resist network failures

Mark Gordon

6

Overview

Introduction

Distributed Shared Memory

COMET Design

Evaluation

Summary

Mark Gordon

7

Distributed Shared Memory

COMET is offloading + DSMOffloading bridges computation disparityDSM provides logically shared address spaceDSM usually applied to cluster environmentsLow latency, high throughputMobile relies on wireless communication

Mark Gordon

8

X=123

X=555

X=555

X=555

X=?

X=?

X=123

DSM (continued)

Conventional DSM (Munin)

X=123

Waited an RTT for a write

Read could take RTT also

Mark Gordon

9

Java Memory Model

Dictates which writes a read can observeSpecifies 'happens-before' partial orderAccess in single thread totally orderedLazy Release Consistency lockingFundamental memory unit is the fieldKnown alignment, known width

Mark Gordon

10

Field DSM

Track dirty fields locally

Need 'happens-before' established?

Transmit dirty fields! (mark fields clean)

Not clear it scales well past two endpoints

Not important to our motivation

Use classic cluster DSM on server

Mark Gordon

11

Overview

Introduction

Distributed Shared Memory

COMET Design

Evaluation

Summary

Mark Gordon

12

VM-synchronization

Used to establish 'happens-before' relation

Directed operation between pusher and puller

Synchronizes

Bytecode sources

Java thread stacks

Java heap

Mark Gordon

13

Bytecode Update (Step 1 of 3)

Operation begins by sending any new code

I loaded file xyz.dex

I have xyz.dex cached

Send xyz.dex

[xyz.dex]

Pusher

Puller

Mark Gordon

14

Stack Update (Step 2 of 3)

Next we send over thread stacks

Thread id: 2

job2::run

pc:5

registers[42, 555, 0]workLoop pc:6 registers[0, [obj:9]]start pc:3 Registers[101, [obj:9]]

Pusher

Puller

nom

Mark Gordon

15

Heap Update (Step 3 of 3)

Finally send over heap update

We send updates to any changed (or new) fieldOnly send updates of 'shared' heap

Pusher

Puller

[obj:2].y = 1

[obj:4].z = [obj:3]

...

Mark Gordon

16

Lock ownership

Annotate with lock ownership flag

Establish 'happens-before' with VM-sync

Mark Gordon

17

Thread Migration

Thread migration trivial

Push VM-syncTransfer lock ownership

Pusher

Puller

Mark Gordon

18

Native Methods

Written in C with bindings for Java

Math.sin(), OSFileSystem.write(), VMThread.currentThread()

Native methods exist to

Access device resources (file system, display, etc)

For performance reasons

To work with existing libraries

Not generally safe to run on either endpoint

Manually white list safe native methods

Mark Gordon

19

Failure Recovery

VM-synchronization is recovery safe

Always leave enough information on clientIf server is lost resume threads running locally!A few caveats (native methods)

Mark Gordon

20

Tau-Scheduler

Τ = 2 * VM-synchronization time

Mark Gordon

21

Implementation

Built from gingerbread CyanogenMod source~5000 lines of C codeJIT not included

Engine.c:offMigrateThread()

offWriteU1(self, OFF_ACTION_MIGRATE); deactivate(self); offThreadWaitForResume(self);

Mark Gordon

22

Overview

Introduction

Distributed Shared Memory

COMET Design

Evaluation

Summary

Mark Gordon

23

Evaluation Setup

Samsung Captivate (1 GHz Hummingbird)2 x 3.16GHz quad core Xeon X5460 cores

Mark Gordon

24

Benchmarks

8 applications from Google Play

Average speed-up of 2.88X on WiFi / 1.28X on 3G

Average energy saving of 1.51X on WiFI / 0.84X on 3G

2 computation benchmark applications

10.4X speed-up w/ WiFi on Linpack

500+X speed-up w/ multi-threaded factoring

Mark Gordon

25

Rhino

Java JavaScript InterpreterRan with SunSpider JavaScript benchmark

Mark Gordon

26

Overview

Introduction

Distributed Shared Memory

COMET Design

Evaluation

Summary

Mark Gordon

27

Summary

Offloading+DSM=COMETImprove computation speedNo programmer effortGeneralize wellResist network failures

Mark Gordon

28

Contributions

Design/Impl. with four simultaneous goals

Fine granularity offloading

Mutli-threading support

Field based DSM coherency

Mark Gordon

29

Questions?

?

Mark Gordon

30

Macrobenchmarks

Mark Gordon

31

Macrobenchmarks (continued)